基于词向量的语言模型已经广泛运用在语言学研究中。现有的研究通过有 监督的方法来探测词义的变化。然而，对一些语言而言，不存在权威的词典例 句作为参考。为了解决这一问题，这篇文章针对四种语言(英语，拉丁语，德语 和瑞典语)，提出了基于深度嵌入词向量将有监督学习和无监督学习结合起来 的方法来探测词义的变化。我们的目标是判断词语是否在两个时间段内获得 了或者失去了某个含义和对词义的变化程度进行量化评分。实验说明我们的 方法在两项任务上是有显著效果的。

Recently, word embeddings have been widely used in linguistic studies. The present studies can only detect lexical semantic change with supervised methods. However, for some languages there is no high-quality example sentences for reference. To address this problem, this paper proposes to combine supervised and unsupervised methods which are both based on deep contextualized embeddings to detect semantic change in four languages - English, German, Latin and Swedish. We aim to determine if a word gain or lose sense during two periods of time and quantify the degree of change in word senses. The experiment shows that our method is effective in binary classification and ranking tasks.